SHACL Generation and Validation¶

5. SHACL generation design¶

5.1 Deterministic Brick-scoped generation¶

For a GPC Brick:

locate the level-4 Brick;
collect level-5 children as Attribute Types;
collect level-6 children as allowed Attribute Values;
emit one sh:NodeShape targeting the Brick;
emit one property shape per populated Attribute Type;
use sh:in for controlled values and sh:maxCount 1 where supported by the demonstrated model;
add cross-category or category-specific GDSN shape modules separately.

5.2 Reusable generation prompt¶

The baseline contained the following reusable implementation prompt, retained here as a technical exhibit with minor editorial normalisation:

You have access to gpc.ttl and, optionally, gdsn.ttl.

Given either:
  (a) a GPC Brick code; or
  (b) a GTIN, in which case first resolve the GTIN to its GPC Brick using
      authoritative trade-item data because the ontology contains schema,
      not per-GTIN product instances;

generate a valid SHACL shape as follows:

1. Locate the Brick in gpc.ttl and verify that it is a level-4 resource.
2. Collect level-5 child classes whose rdfs:subClassOf is the Brick.
3. For each Attribute Type, collect level-6 child classes whose
   rdfs:subClassOf is that Attribute Type.
4. Emit one sh:NodeShape targeting the Brick and one sh:property per
   Attribute Type, including sh:path, sh:name, sh:in and sh:maxCount 1.
5. Declare every prefix required for standalone parsing.
6. Where gdsn.ttl is available, emit the selected GDSN shape as a separate
   sh:NodeShape, using real datatypes, classes, multiplicities and code-list
   references from the generated ontology.
7. Validate the output with pySHACL against at least one conforming and one
   deliberately broken instance, and report the result.

5.3 Reference generator excerpt¶

import re


def load_gpc(path="gpc.ttl"):
    text = open(path, encoding="utf-8").read()
    records = text.split("\n\n")

    def parse(record):
        item = {}
        match = re.match(r"gpc:(\S+)", record)
        if match:
            item["id"] = match.group(1)
        match = re.search(r"gpc:level (\d+)", record)
        if match:
            item["level"] = int(match.group(1))
        match = re.search(r'rdfs:label "([^"]*)"', record)
        if match:
            item["label"] = match.group(1)
        match = re.search(r"rdfs:subClassOf gpc:(\S+)", record)
        if match:
            item["parent"] = match.group(1)
        return item

    parsed = [parse(record) for record in records if "a owl:Class" in record]
    by_id = {item["id"]: item for item in parsed if "id" in item}
    children = {}
    for item in parsed:
        if "parent" in item:
            children.setdefault(item["parent"], []).append(item["id"])
    return by_id, children


def gpc_shape_for_brick(brick_code, by_id, children, shape_ns="gpc-shapes:"):
    brick = by_id.get(brick_code)
    if brick is None:
        raise ValueError(f"Brick {brick_code} not found in GPC ontology")

    attribute_types = [
        by_id[child_id]
        for child_id in children.get(brick_code, [])
        if by_id[child_id].get("level") == 5
    ]

    slug = lambda label: re.sub(r"[^A-Za-z0-9]+", "", label)
    shape_name = f"{shape_ns}{slug(brick['label'])}Shape"
    lines = [
        f"{shape_name} a sh:NodeShape ;",
        f"    sh:targetClass gpc:{brick_code} ;  # {brick['label']}",
    ]

    for attribute_type in attribute_types:
        values = [
            by_id[child_id]["label"]
            for child_id in children.get(attribute_type["id"], [])
            if by_id[child_id].get("level") == 6
        ]
        if not values:
            continue
        in_list = " ".join(f'"{value}"' for value in values)
        lines += [
            "    sh:property [",
            f'        sh:path gpc:{attribute_type["id"]} ; '
            f'sh:name "{attribute_type["label"]}" ;',
            f"        sh:in ( {in_list} ) ;",
            "        sh:maxCount 1 ;",
            "    ] ;",
        ]

    lines[-1] = lines[-1].rstrip(" ;") + " ."
    return shape_name, "\n".join(lines)

The excerpt demonstrates the deterministic pattern but is not a complete production parser. It relies on source formatting conventions and should be replaced or hardened with RDF parsing where appropriate.

6. SHACL validation evidence¶

6.1 Corrected illustrative SHACL¶

The early source fragments were corrected to provide standalone prefixes and independently targeted node shapes. The following compact form preserves the demonstrated structure:

@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix gpc: <gpc:> .
@prefix gdsn: <gdsn:> .
@prefix gpc-shapes: <https://example.org/gpc-shapes#> .

gpc-shapes:CheeseFrozenShape a sh:NodeShape ;
    sh:targetClass gpc:10000030 ;
    sh:property [
        sh:path gpc:20000192 ;
        sh:name "Firmness of Cheese" ;
        sh:in ( "HARD" "SOFT" "FIRM/SEMI-HARD" "EXTRA HARD" ) ;
        sh:maxCount 1 ;
    ] ;
    sh:property [
        sh:path gpc:20000031 ;
        sh:name "Type of Cheese" ;
        sh:in ( "GOUDA" "CHEDDAR" "BRIE" "FETA" "MOZZARELLA" "EDAM" "PARMESAN" "GRUYERE" ) ;
        sh:maxCount 1 ;
    ] .

gpc-shapes:TradeItemGDSNShape a sh:NodeShape ;
    sh:targetClass gdsn:TradeItem ;
    sh:property [
        sh:path gdsn:gtin ;
        sh:datatype xsd:string ;
        sh:minCount 1 ;
        sh:maxCount 1 ;
    ] ;
    sh:property [
        sh:path gdsn:allergenRelatedInformation ;
        sh:maxCount 1 ;
        sh:node [
            a sh:NodeShape ;
            sh:property [
                sh:path gdsn:allergen ;
                sh:node [
                    a sh:NodeShape ;
                    sh:property [
                        sh:path gdsn:allergenTypeCode ;
                        sh:in ( "ML" "MFD" "BA" "SU" "PHD" "TUR" ) ;
                        sh:minCount 1 ;
                        sh:maxCount 1 ;
                    ] ;
                    sh:property [
                        sh:path gdsn:levelOfContainmentCode ;
                        sh:minCount 1 ;
                        sh:maxCount 1 ;
                    ] ;
                ] ;
            ] ;
        ] ;
    ] ;
    sh:property [
        sh:path gdsn:nutrientHeader ;
        sh:maxCount 1 ;
        sh:node [
            a sh:NodeShape ;
            sh:property [
                sh:path gdsn:nutrientDetail ;
                sh:node [
                    a sh:NodeShape ;
                    sh:property [
                        sh:path gdsn:nutrientTypeCode ;
                        sh:in ( "ENERC" "FAT" "PRO-" "NA" "CA" ) ;
                        sh:minCount 1 ;
                        sh:maxCount 1 ;
                    ] ;
                    sh:property [
                        sh:path gdsn:quantityContained ;
                        sh:datatype xsd:decimal ;
                    ] ;
                ] ;
            ] ;
        ] ;
    ] .

This is an illustrative logical-name shape. The real generated artefacts use source-ID-based CURIEs and, for several associations, sh:class references rather than anonymous nested sh:node blocks.

6.2 Reported `pySHACL` results¶

Test instance	Reported result
Conforming Cheese record with GTIN, permitted GPC values, allergen code plus containment level, and permitted nutrient code	`Conforms: True`
Broken record with firmness `SQUISHY`, nutrient code `PROTEIN_XYZ`, and missing containment level	`Conforms: False`

The broken record produced three reported violations:

two sh:InConstraintComponent violations;
one sh:MinCountConstraintComponent violation.

The tests demonstrate that category-inappropriate values and missing structured subfields can be detected independently.

6.3 Generalisation evidence¶

The source baseline reported successful shape generation and parse/validation checks for:

Brick	Attribute Types reported	Result
`10000030` Cheese (Frozen)	6	Parsed; conforming test passed.
`10001198` Smartphones	2	Parsed; conforming test passed.
`10000159` Beer	9	Parsed; conforming test passed.

A fourth supplied shape, Sugar/Sugar Substitutes (Shelf Stable), contained one Attribute Type and the same selected GDSN shape as the other examples.

6.4 Separation of shape and instance¶

A material correction from the source history is retained here:

the Product Holon is the product-instance graph;
the SHACL shape is a separately versioned validation contract;
the GDSN and GPC ontology modules provide semantic definitions;
the SSSOM mapping set governs semantic correspondences;
the bridge package implements approved mappings;
the Web Vocabulary/schema.org graph is a derived publication view.

Conflating these artefacts prevents independent versioning, provenance and conformance testing.

6.5 Reusable code-list strategy¶

Large code lists should not be unnecessarily repeated in every Brick shape. The evidence identified 218 allergen types and 1,080 nutrient types in the inspected GDSN model. A production SHACL design should consider reusable code-list resources, referenced shapes or generated modules so that one code-list change does not require duplicating a large sh:in list across thousands of Brick files.

6.6 Real generated shape excerpt¶

The following excerpt preserves the opaque source-ID-based CURIEs from the supplied Cheese (Frozen) shape. Enumerations are abbreviated only where the source baseline itself abbreviated them.

@prefix gdsn: <gdsn:> .
@prefix gpc: <gpc:> .
@prefix gpc-shapes: <gpc-shapes:> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

gpc-shapes:Brick10000030Shape a sh:NodeShape ;
    rdfs:label "Cheese (Frozen) GPC Brick shape" ;
    sh:name "Cheese (Frozen) GPC Brick shape" ;
    sh:property [
        sh:in ( "GOUDA" "CHEDDAR" "BRIE" "FETA" "MOZZARELLA" "EDAM" "PARMESAN" "GRUYERE" "..." ) ;
        sh:maxCount 1 ;
        sh:name "Type of Cheese" ;
        sh:path gpc:20000031
    ], [
        sh:in ( "EXTRA HARD" "FIRM/SEMI-HARD" "HARD" "SOFT" ) ;
        sh:maxCount 1 ;
        sh:name "Firmness of Cheese" ;
        sh:path gpc:20000192
    ], [
        sh:in ( "BARBEQUE" "DIRECT CONSUMPTION" "FONDUE" "SMOKING" ) ;
        sh:maxCount 1 ;
        sh:name "Intended Use of Cheese" ;
        sh:path gpc:20003080
    ], [
        sh:in ( "SEMI-HARD CHEESE" "SEMI-SOFT CHEESE" "SOFT / SOFT RIPENED CHEESE" "..." ) ;
        sh:maxCount 1 ;
        sh:name "Kind of Cheese" ;
        sh:path gpc:20003081
    ], [
        sh:in ( "EXTRA EXTRA SHARP" "EXTRA SHARP" "SHARP" ) ;
        sh:maxCount 1 ;
        sh:name "Sharpness of Cheese" ;
        sh:path gpc:20002867
    ], [
        sh:in ( "AUSTRIA - TYROL/VORARLBERG/CARINTHIA" "AUSTRIA - UPPER/LOWER AUSTRIA" ) ;
        sh:maxCount 1 ;
        sh:name "Origin of Cheese" ;
        sh:path gpc:20000505
    ] ;
    sh:targetClass gpc:10000030 .

gpc-shapes:TradeItemGDSNShape a sh:NodeShape ;
    rdfs:comment "Generated from selected GDSN properties. Source property domains are preserved as sh:description because not all selected properties are direct TradeItem properties in the generated ontology." ;
    sh:name "Cross-category GDSN TradeItem shape" ;
    sh:property [
        sh:datatype gdsn:c1450 ;
        sh:maxCount 1 ;
        sh:minCount 0 ;
        sh:name "gtin" ;
        sh:path gdsn:a-1292425203
    ], [
        sh:datatype gdsn:c1481031847 ;
        sh:minCount 0 ;
        sh:name "ingredientStatement" ;
        sh:path gdsn:a812659001
    ], [
        sh:datatype xsd:float ;
        sh:maxCount 1 ;
        sh:minCount 0 ;
        sh:name "packagingRecycledContentRatio" ;
        sh:path gdsn:a212299678
    ], [
        sh:class gdsn:c2147348056 ;
        sh:minCount 0 ;
        sh:name "claimDetail" ;
        sh:path gdsn:assoc_45067187_2147348056_4
    ], [
        sh:class gdsn:c1672558995 ;
        sh:minCount 0 ;
        sh:name "allergen" ;
        sh:path gdsn:assoc_-474206270_1672558995_1
    ], [
        sh:class gdsn:c1322 ;
        sh:minCount 0 ;
        sh:name "nutrientDetail" ;
        sh:path gdsn:assoc_1336934671_1322_1
    ], [
        sh:class gdsn:c-474206270 ;
        sh:minCount 0 ;
        sh:name "allergenRelatedInformation" ;
        sh:path gdsn:assoc_-1935341322_-474206270_1
    ], [
        sh:class gdsn:c1336934671 ;
        sh:minCount 0 ;
        sh:name "nutrientHeader" ;
        sh:path gdsn:assoc_1340910615_1336934671_2
    ] ;
    sh:targetClass gdsn:c863999331 .

The shape’s own comment records the flattening decision: selected properties from several source domains are validated against a TradeItem-targeted shape, while the original domains remain annotations. This decision must be aligned with the Product Holon construction convention.